Data Centre & AI Infrastructure Buildout
The global AI boom has triggered a capital expenditure surge in data centre infrastructure that is reshaping demand across multiple supply chains simultaneously. Hyperscalers (Microsoft, Google, Amazon, Meta) and sovereign AI infrastructure programmes are committing hundreds of billions of dollars to training clusters, inference capacity, and supporting power and cooling infrastructure. This buildout is creating structural demand for copper, steel, power equipment, land, and skilled construction and operations labour at a scale and speed that is straining multiple upstream supply chains.
Chain-Level Impact
How this trend is affecting each named supply chain — direction of pressure and strategic significance.
Data Centre Supply Chain
The data centre buildout is the defining demand event for the entire data centre supply chain.
Global data centre capex is forecast to reach $500B+ by 2027, growing at 20–30% annually. This is creating unprecedented demand for all supply chain nodes: site development, structural construction, mechanical and electrical systems, networking, and compute hardware.
Copper Supply Chain
Data centre power infrastructure, busbars, cables, and liquid cooling loops are driving substantial copper demand.
A large AI hyperscale data centre (500MW) requires thousands of tonnes of copper for power distribution, grounding, liquid cooling, and interconnect cabling. The density of AI workloads (GPU clusters) increases copper intensity per MW significantly compared to CPU-based facilities.
Steel Supply Chain
Data centre structural steel for buildings, server racks, and support infrastructure is growing demand.
Each large hyperscale campus requires tens of thousands of tonnes of structural steel for the building envelope, raised floor systems, and server rack structures. Speed of construction is also driving demand for prefabricated modular data centre structures (pre-engineered steel buildings).
Semiconductor Supply Chain
AI accelerators (GPUs, TPUs, custom ASICs) are the highest-value component driving the data centre buildout.
NVIDIA H100/H200/B100 GPU clusters are the primary AI training hardware. Each rack of H100s costs $400K–$600K and requires specialised power and cooling. Custom silicon from Google (TPUs), Amazon (Trainium/Inferentia), and Microsoft (Maia) is creating additional demand for advanced TSMC capacity.
Battery Supply Chain
Data centres require large UPS battery systems for power conditioning and backup — but AI energy demand competes with grid batteries.
Data centre UPS systems (typically lead-acid for legacy, increasingly lithium-ion for new builds) require significant battery capacity per facility. However, the massive grid power draw of AI data centres is straining the same electricity infrastructure that battery storage is meant to support.
Winners & Losers
Industries facing headwinds (cost, risk, constraint) and tailwinds (demand, opportunity, advantage) from this trend.
↑ Tailwinds (5)
Data Processing, Hosting and Related Activities
Hyperscale cloud providers, co-location operators, and AI-specialised data centre operators are all experiencing strong revenue growth. The cloud IaaS market is growing at 25%+ annually driven by AI workload demand.
Manufacture of Basic Precious and Other Non-Ferrous Metals
Copper is the primary metal beneficiary of data centre buildout. Aluminium (used in cooling systems and enclosures) is also in strong demand. Smelters and rolled product manufacturers serving the electrical equipment segment are seeing order backlogs extend significantly.
Manufacture of Basic Iron and Steel
Data centre construction is a meaningful demand driver for structural steel in major buildout markets (US, Europe). Prefabricated data centre structures and modular UPS enclosures use cold-rolled steel sheet extensively.
Architectural and Engineering Activities and Related Technical Consultancy
Mission-critical facility engineers (MEP, structural, civil) with data centre experience are in extreme shortage. Engineering firms specialising in hyperscale data centre design are booked 18–24 months forward. Data centre engineering is one of the highest-growth specialisms in the AEC sector.
Computer Programming Activities
Software companies serving AI infrastructure (DCIM platforms, orchestration tools, monitoring software) are capturing a share of the data centre buildout. DCIM providers such as Nlyte, Sunbird, and Hyperview are growing rapidly.
Which Strategic Pillars Are Activated
The GTIAS pillar attributes most activated by this trend — signalling which parts of an industry's risk profile are most likely to deteriorate.
Infrastructure
Power availability has emerged as the binding constraint on data centre development in most major markets. Grid interconnect queues run to 5–10 years in some US and UK markets. Hyperscalers are directly funding power infrastructure — including nuclear power purchase agreements and gas peaker plants — to secure guaranteed electricity supply for AI workloads.
Market Dynamics
AI infrastructure is a winner-take-most market where compute density, power efficiency, and geographic location (latency, energy cost, regulatory jurisdiction) determine competitive position. Co-location providers and hyperscalers are competing aggressively for prime data centre sites globally.
Supply Chain
Data centre supply chains — GPUs, networking switches, power distribution units, liquid cooling systems, structural steel, and power transformers — are simultaneously supply-constrained. Lead times for critical equipment (HV transformers: 2–3 years; NVIDIA GPUs: allocation-constrained) are creating project delivery risk.
Resource Procurement
Power transformer procurement is the most acute supply chain bottleneck in the data centre buildout. Transformer manufacturers (ABB, Siemens, Eaton, Hitachi) are booking 24–36 months forward. New transformer manufacturing capacity takes years to build. This is a structural constraint on data centre delivery timelines.
What This Means for Strategy
Power is the strategic moat for data centre operators. Companies that have secured long-term power purchase agreements with low-carbon generators, or that control their own generation assets (solar, nuclear PPAs), have a structural advantage over operators dependent on utility grid allocation.
The AI infrastructure supply chain (transformers, liquid cooling, power distribution, GPUs) is operating with multi-year lead times. Hyperscalers are locking up supply 2–3 years forward; smaller operators face allocation risk that threatens project delivery timelines.
Geographic diversification of data centre capacity is increasing as single-market concentration risk becomes apparent. EU data sovereignty, tax jurisdiction, and regulatory considerations are driving European hyperscale investment; Middle East and Southeast Asia are growing as alternative hub markets.